Background: Let $(Y, \mathcal{B},\mu,T)$ be an ergodic probability system and let $G$ be a compact metrizable group with compact subgroup $H$. Given a measurable map $\rho:Y \to G$. We may define the skew product system $W_{H,\rho}=(Y \times G/H, \mu \times m_{G/H}, T_{\rho})$ where the transformation is given by $$ T_{\rho}(y,u)=(Ty, \rho(y)u)$$ and by $m_{G/H}$ we mean the pushforward onto $G/H$ of the Haar measure on $G$. Note that for $g \in G$ we have a measure preserving map $g:Y \times G/H \to Y \times G/H$ given by $$g.(y,u)=(y,gu)$$ and thus $G$ acts on the Hilbert Space $L^2(W_{H,\rho})$.
Question: If the system $(Y \times G, \mu \times m_G, T_{\rho})$ is ergodic, then is it true that for any eigenfunction $e: Y \times G/H \to \mathbb{C}$ of $W_{H,\rho}=(Y \times G/H, \mu \times m_{G/H}, T_{\rho})$ we have that $e \circ [g_1,g_2]=e$ for $g_1,g_2 \in G$? i.e all eigenfunctions are fixed by the commutator subgroup $[G,G]$?
Motivation: It seems to me that this is suggested in the proof of Lemma 6.1 in Nonconventional ergodic averages and nilmanifolds by Host-Kra (available to public here http://www.math.northwestern.edu/~kra/papers/convnil.pdf)